I have to estimate a simultaneous equation system (regressions) with about 11 equations and 21 variables.
I estimated by 2SLS method in econometric and now I want to estimate it with genetic algorithm. Can any one help me on how to start? Is there any paper related to this(estimation of simultaneous equation with GA)? Do you know what should my fitness function be?
Thanks for any help.
Take a look at Grammatical Evolution. You provide a list of the variables, available functions and a fitness evaluation function. The program will use genetic algorithms to produce genomes that map to different programs to evaluate. Using the fitness function to minimize error, it will evolve a program to solve the problem.
One possible implementation to use is GEVA which is implemented in JAVA.
If you don't know how many equations are involved in a complex dataset, you will probably want to look into using meta-grammars to define the available functions and variables. I'd need to know more about the specific problem to give any better advise.
Very similar question here.
This paper might help, it defines a fitness function that should approximate the solution but keep in mind that's not the only option.